CN113434692B - Method, system and equipment for constructing graphic neural network model and recommending diagnosis and treatment scheme - Google Patents
Method, system and equipment for constructing graphic neural network model and recommending diagnosis and treatment scheme Download PDFInfo
- Publication number
- CN113434692B CN113434692B CN202110690423.0A CN202110690423A CN113434692B CN 113434692 B CN113434692 B CN 113434692B CN 202110690423 A CN202110690423 A CN 202110690423A CN 113434692 B CN113434692 B CN 113434692B
- Authority
- CN
- China
- Prior art keywords
- entity
- data
- diagnosis
- relation
- patient
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/10—Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
Abstract
The invention discloses a method, a system and equipment for constructing a graph neural network model and recommending a diagnosis and treatment scheme. The method for constructing the graph neural network model comprises the following steps: acquiring historical diagnosis and treatment data of a plurality of patients; according to the predefined relation between the entity and every two entities in the historical diagnosis and treatment data, constructing a knowledge graph triplet to establish a knowledge graph; and constructing a graph neural network model according to the knowledge graph. According to the graph neural network model construction and diagnosis and treatment scheme recommendation method, system and equipment, rich semantic association among entities can be fully mined, entity types and relationship types are reasonably expanded, recommendation accuracy is improved, and recommendation diversity is increased.
Description
Technical Field
The invention relates to the medical field, in particular to a method, a system and equipment for constructing a graphic neural network model and recommending a diagnosis and treatment scheme.
Background
In recent years, in the traditional medical industry, artificial intelligence gradually plays an advantage and a unique role in a plurality of links such as disease auxiliary diagnosis and treatment, personal health management, gene medicine research and development, hospital intelligent management and the like. During a disease visit, hospitals maintain a large number of information records (including patient basic information, diagnostic information, past history, and examination records) of related visits, as well as medical regimens (surgery, medication, adjuvant therapy, etc.) for patient conditions. The large amount of history records implies a large amount of information, and if the artificial intelligence technology can be well applied, valuable information can be mined from the history records, so that the aim of disease auxiliary diagnosis and treatment is fulfilled.
The traditional machine learning diagnosis and treatment scheme recommendation algorithm uses the history diagnosis and treatment record information of patients, takes some key factors in the diagnosis and treatment process as characteristics to be combed out, and trains a recommendation algorithm model according to the characteristics. In the diagnosis and treatment scheme recommendation method based on deep learning, a similar medical record model is trained by using medical record samples and corresponding labeling labels, and then the similar medical record model is matched to achieve the recommendation of the corresponding diagnosis and treatment scheme.
In summary, the development of the above method has achieved a certain effect in different scenes, but in some complex scenes, both the machine learning algorithm and the deep learning algorithm can only dig out shallow semantic information and related information, and the related information of deeper layers cannot be fully utilized.
Disclosure of Invention
The present invention aims to solve at least one of the technical problems in the related art to some extent. Therefore, an object of the present invention is to provide a method for constructing a graph neural network model, which has the advantages that the knowledge graph can fully mine rich semantic association between entities, and can fully utilize deeper relationship information in the historical diagnosis and treatment data of a patient, so as to improve recommendation accuracy.
The invention aims to provide a graph neural network model construction method, which has the advantages that the knowledge graph has interpretability and the relationship types in the knowledge graph are rich, so that the entity types and the relationship types can be reasonably expanded, and the recommendation diversity is increased.
The invention aims to provide a graph neural network model construction method, which is further advantageous in neighborhood aggregation, capturing and storing local adjacent structures in each entity.
The invention aims to provide a diagnosis and treatment scheme recommendation method which is high in recommendation accuracy and comprehensive in recommendation information.
The invention provides a graph neural network model construction method, which comprises the following steps:
acquiring historical diagnosis and treatment data of a plurality of patients;
according to the entity in the historical diagnosis and treatment data and the predefined relation between every two entities, a knowledge graph triplet is constructed to establish the knowledge graph; and constructing a graph neural network model according to the knowledge graph.
The graph neural network model construction method of the invention has the advantages that: the rich semantic association among knowledge graph entities helps to mine the association therein, thereby improving the accuracy of the constructed graph neural network model. The relationship types in the knowledge graph are rich, so that the service requirements can be reasonably expanded, and the recommendation diversity is increased. The knowledge graph has good explanatory property, can clearly understand the relation among the entities, and is convenient for constructing the graph neural network model.
In addition, the construction method of the graph neural network model according to the invention can also have the following additional technical characteristics:
in some embodiments of the present invention, the historical diagnosis and treatment data includes structured data and unstructured data, and the construction of the knowledge-graph triples includes the following steps: directly extracting entities and relations in the structured data; extracting entities in the unstructured data according to an entity extraction model to generate entity data; extracting the relation between every two entities in the entity data according to a relation extraction model; and constructing the knowledge-graph triples according to the entity and the relation.
In some embodiments of the present invention, extracting entities in the unstructured data using an entity extraction model comprises the steps of:
1) Obtaining the unstructured data;
2) Labeling entity types of entities in the unstructured data to generate entity type training data;
3) Training a preset entity learning model according to the entity type training data to generate the entity extraction model;
4) Predicting entity types of entities in the unstructured data according to the entity extraction model to generate predicted data, wherein the predicted unstructured data is generated by deleting unstructured data corresponding to the entity type training data in the unstructured data;
5) Checking whether the entity in the predicted data is correctly corresponding to the entity type, and adding the correctly corresponding predicted data into the entity type training data to generate updated entity type training data;
6) Determining whether entity types of all entities in the prediction data are correctly corresponding; if not, replacing the entity type training data by using the updated entity type training data, and repeating the steps 3) -5); and if so, extracting the entity in the last generated updated entity type training data and the predicted data to generate the entity data.
In some embodiments of the invention, extracting the relationship from the entity data comprises the steps of:
7) Dividing sentences in the entity data according to periods;
8) Finding out a possible combination set of every two entities in sentences of the clause;
9) Marking the relation between every two entities in part of the entity data according to the combined set to generate relation training data;
10 Training a preset relationship learning model according to the relationship training data to generate the relationship extraction model;
11 Predicting the relation between every two entities in predicted entity data according to the relation extraction model to generate predicted relation data, wherein the predicted entity data is generated by deleting entity data corresponding to the relation training data in the entity data;
12 Checking whether the relation between every two entities in the predicted relation data is correctly corresponding, and adding the correctly corresponding predicted relation data into the relation training data to generate updated relation training data;
13 Determining whether the relation between every two entities in the predicted relation data is correctly corresponding; if not, replacing the relation training data by adopting updated relation training data, and repeating the steps 10) -12); and if so, extracting the relationship between the updated relationship training data and the predicted relationship data which are generated last time.
In some embodiments of the present invention, constructing a graph neural network model from the knowledge-graph includes the steps of:
constructing a patient set, an entity set and a relation set;
constructing a patient-entity matrix according to the interaction conditions of each patient and each entity in the patient set and the entity set;
constructing a patient feature vector according to the patient-entity matrix;
non-linearly transforming the patient feature vector into a D-dimensional patient feature vector, D being a positive integer;
converting each entity in the entity set into a D-dimensional vector to construct an entity vector matrix;
converting each relationship in the relationship set into a D-dimensional vector to construct a relationship vector matrix;
Constructing an entity-entity neighborhood matrix according to the entity set and the adjacent situation of every two entities in the knowledge graph;
constructing an entity-relationship neighborhood matrix according to the relationship set, the entity set and the adjacent conditions of the entity and the relationship in the knowledge graph;
fusing the entity-entity neighborhood matrix and the entity vector matrix to generate an entity neighborhood matrix;
fusing the entity-relationship neighborhood matrix and the relationship vector matrix to generate a relationship neighborhood matrix;
performing inner product and one-dimensional summation on the relation neighborhood matrix and the D-dimensional patient feature vector to generate a patient-relation score;
performing inner product on the patient-relation score and the entity neighborhood matrix to generate a patient neighborhood matrix;
adding the patient neighborhood matrix and the entity neighborhood matrix to generate an updated patient neighborhood matrix;
and carrying out inner product and one-dimensional summation on the updated patient neighborhood matrix and the D-dimensional patient feature vector to generate a diagnosis and treatment scheme vector so as to construct a completed graph neural network model.
The invention also provides a graph neural network model construction system, which comprises: the acquisition module is used for acquiring historical diagnosis and treatment data; the first construction module is used for constructing a knowledge spectrum triplet according to the entity in the historical diagnosis and treatment data and the predefined relation between every two entities so as to establish the knowledge spectrum; and the second construction module is used for constructing a graph neural network model according to the knowledge graph.
In addition, the graph neural network model according to the invention can also have the following additional technical characteristics:
in some embodiments of the present invention, the historical diagnosis and treatment data includes structured data and unstructured data, and the graph neural network model building system further includes: the first extraction module is used for directly extracting entities and relations in the structured data; the second extraction module is used for extracting entities in the unstructured data according to an entity extraction model so as to generate entity data; and extracting the relation between every two entities in the entity data according to a relation extraction model.
The invention also provides a recommendation method of the diagnosis and treatment scheme, which comprises the following steps:
acquiring diagnosis and treatment entity data of a plurality of patients;
constructing diagnosis and treatment knowledge map triplets according to diagnosis and treatment entities in the diagnosis and treatment entity data and a predefined diagnosis and treatment relationship between every two diagnosis and treatment entities so as to establish a plurality of diagnosis and treatment knowledge maps;
processing a plurality of said diagnosis and treatment knowledge maps according to a graph neural network model constructed according to a construction method as described above to generate a plurality of diagnosis and treatment plan vectors;
acquiring medical entity data of a patient to be recommended;
According to the medical entity data, medical knowledge graph triplets are constructed according to the medical relation predefined between the medical entity and every two medical entities so as to establish a medical knowledge graph;
processing the medical knowledge graph according to the graph neural network model to generate a medical vector to be recommended;
calculating the similarity between the medical vector to be recommended and each diagnosis and treatment scheme vector; the method comprises the steps of,
and recommending a diagnosis and treatment scheme corresponding to the diagnosis and treatment scheme vector with the similarity value meeting the preset condition according to the similarity.
The invention also provides a system for recommending the diagnosis and treatment scheme, which comprises: the acquisition module is used for acquiring diagnosis and treatment entity data of a plurality of patients and medical entity data of patients to be recommended; the third construction module is used for constructing diagnosis and treatment knowledge map triplets according to diagnosis and treatment entities in the diagnosis and treatment entity data and the diagnosis and treatment relation predefined between every two diagnosis and treatment entities so as to establish diagnosis and treatment knowledge maps; the medical knowledge graph triad is used for building a medical knowledge graph according to the medical entity in the medical entity data and the predefined medical relation between every two medical entities so as to build the medical knowledge graph; the processing module is used for processing a plurality of diagnosis and treatment knowledge maps according to the graph neural network model so as to generate a plurality of diagnosis and treatment scheme vectors; the medical knowledge graph is used for processing the medical knowledge graph according to the graph neural network model to generate a medical vector to be recommended; the calculation module is used for calculating the similarity between the medical vector to be recommended and each diagnosis and treatment scheme vector; and the selecting module is used for selecting a diagnosis and treatment scheme corresponding to the diagnosis and treatment scheme vector with the similarity value meeting the preset condition according to the similarity.
The invention also provides an electronic device, which comprises a processor and a memory; wherein the processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, for implementing the graph neural network model construction method as described above.
The present invention also proposes a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the graph neural network model building method as described above.
Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.
Detailed Description
Embodiments of the present invention are described in detail below. The following examples are illustrative and are intended to be illustrative of the invention and are not to be construed as limiting the invention.
A method for constructing a graph neural network model comprises the following steps:
s10: historical diagnosis and treatment data of a plurality of patients are obtained. For example, the historical diagnosis and treatment data includes data such as basic information, symptoms, past history, examination information, and treatment information of a patient suffering from a disease. The basic information includes the age, sex, and BMI value of the patient, etc. The past history includes the history of smoking and drinking of the patient, and the symptoms include subjective abnormal sensations or some objective pathological changes of the patient suffering from the disease, such as fever, cough, headache, and the like. The examination information includes examination items of the body by the patient for determining the disease, and body index data obtained by the examination items, and the like, for example, examination items including electrocardiogram, CT chart, blood lipid examination, blood glucose examination, and the like, heart function-related index data, blood glucose-related index data, blood lipid-related index data, and the like, obtained by the examination items. The treatment information includes medicines taken by the patient for curing diseases and operations performed.
S20: and constructing a knowledge graph triplet according to the entity in the historical diagnosis and treatment data and the predefined relation between every two entities so as to establish a knowledge graph. For example, the historical diagnosis and treatment data includes basic information, symptoms, past history, examination items, various physical indexes acquired by examination items, physical information such as operations and medicines, and the like. Relationships between entities are predefined, such as a relationship between a "disease" entity and a "disease" entity is defined as "complications", a relationship between a "disease" entity and a "drug" entity is defined as "drug treatment", a relationship between a "disease" entity and a "test item" entity is defined as "test", "a relationship between a" test "entity and a" surgical name "entity is defined as" surgical treatment ", a relationship between a" disease "entity and a" lifestyle "entity is defined as" personal history "," a relationship between a "disease" entity and "other treatment" entities is defined as "adjuvant treatment", and so forth. And constructing a knowledge graph triplet according to the entity and the predefined relation between every two entities so as to establish a knowledge graph. For example, a knowledge-graph triplet is constructed that is disease-complication-disease based on the "disease" entity, and the "complication" relationship between the two entities described above. According to the above-mentioned "disease" entity, "medicine" entity and the relationship "medicine treatment" between two, the constructed knowledge-graph triad is disease-medicine treatment-medicine. According to the 'disease' entity, the 'inspection item' entity and the 'inspection' relationship between the two entities, the constructed knowledge-graph triplets are disease-inspection items. According to the disease entity, the life habit entity and the personal history of the relationship between the two entities, the constructed knowledge graph triplets are disease-personal history-life habit.
S30: and constructing a graph neural network model according to the knowledge graph.
According to the method, the knowledge graph is formed by adopting the historical diagnosis and treatment data of a plurality of patients, the knowledge graph can fully mine rich semantic association among entities of the historical diagnosis and treatment data, deep information among the entities of the historical diagnosis and treatment data is fully utilized, and the recommendation accuracy in the follow-up recommendation of the diagnosis and treatment scheme is improved. Meanwhile, entity types and relationship types can be reasonably expanded, and recommendation diversity is increased.
Specifically, the historical diagnosis and treatment data comprises structured data and unstructured data, and the construction of the knowledge graph triples comprises the following steps:
s21: entities and relationships in the structured data are directly extracted.
S22: and extracting the entity in the unstructured data according to the entity extraction model to generate entity data.
S23: and extracting the relation between every two entities in the entity data according to the relation extraction model.
S24: and constructing a knowledge graph triplet according to the entity and the relation.
Further, step S22 includes the steps of:
1) Unstructured data is acquired.
2) The entity types of the entities in the portion of unstructured data are annotated to generate entity type training data.
3) Training a preset entity learning model according to the entity type training data to generate an entity extraction model.
4) Predicting entity types of entities in the unstructured data according to the entity extraction model to generate predicted data, wherein the predicted unstructured data is generated by deleting unstructured data corresponding to entity type training data in the unstructured data.
5) And checking whether the entity in the predicted data is correctly corresponding to the entity type, and adding the correctly corresponding predicted data into the entity type training data to generate updated entity type training data.
6) Determining whether all entities and entity types in the predicted data are correctly corresponding; if not, replacing the entity type training data by the updated entity type training data, and repeating the steps 3) -5); if yes, extracting the entity in the last generated updated entity type training data and the predicted data to generate entity data.
The entity extraction method of the invention continuously and alternately operates the training and checking of the entity extraction model so as to achieve the aim of accurately marking the entities in unstructured data and improving the accuracy of entity extraction.
Further, step S23 includes the steps of:
7) Dividing sentences in the entity data into sentences according to periods;
8) Finding out a possible combination set of every two entities in sentences of the clause;
9) Marking the relation between every two entities in part of entity data according to the combined set to generate relation training data;
10 Training a preset relationship learning model according to the relationship training data to generate a relationship extraction model;
11 Predicting the relation between every two entities in the predicted entity data according to the relation extraction model to generate predicted relation data, wherein the predicted entity data is generated by deleting entity data corresponding to relation training data in the entity data;
12 Checking whether the relation between every two entities in the predicted relation data is correctly corresponding, and adding the correctly corresponding predicted relation data into relation training data to generate updated relation training data;
13 Determining whether the relation between every two entities in the predicted relation data is correctly corresponding; if not, replacing the relationship training data by adopting the updated relationship training data, and repeating the steps 10) -12); if yes, extracting the relationship in the updated relationship training data and the predicted relationship data which are generated for the last time.
The relation extraction method of the invention continuously and alternately operates the training and the checking of the relation extraction model so as to achieve the aim of accurately marking the relation between the entities of unstructured data and improving the accuracy of relation extraction.
Specifically, step S30 includes the steps of:
s31: a patient set, an entity set, and a relationship set are constructed.
For example, from the acquired diagnosis and treatment data of M patients, a patient set U, u= { U1, U2, …, uM }, where U represents each patient, is constructed.
From all the entities that M patients have, i.e. N entities, a set of entities V, v= { V1, V2, …, vN }, where V represents each entity, is constructed. For example, a certain drug (nitroglycerin injection), a certain disease (coronary heart disease) and a certain examination (electrocardiogram) all constitute one entity.
From all the relationships that M patients have, i.e., P relationships, a relationship set W, w= { W1, W2, …, wP }, where W represents each relationship, is constructed. For example, the relationship "surgical treatment" between the entities of "acute lower wall posterior wall right ventricular myocardial infarction" and "coronary stent implantation" is one relationship.
S32: and constructing M-N patient-entity matrixes according to the interaction conditions of each patient and each entity in the patient set U and the entity set V. For example, in the patient entity interaction matrix, it is determined whether there is an interaction relationship between the patient and the entity, and if there is an interaction, it is 1, otherwise it is marked as 0. Each row N-dimension in the patient-entity matrix represents a patient feature vector. For example, one row of data of the patient-entity matrix is extracted to form the patient feature vector.
S33: the patient feature vector is converted to a D-dimensional patient feature vector by a nonlinear transformation. D is a positive integer. The dimensions of the patient feature vector are [1, d ], i.e., the dimensions in which the marker in the patient feature vector marked 0 is eliminated.
S34: each entity in the set of entities V is converted into a D-dimensional vector to construct an entity vector matrix.
S35: and constructing an N-N entity-entity neighborhood matrix according to the entity set V and the adjacent condition of every two entities in the knowledge graph. For example, for one entity V in the entity set V, it is determined whether the entity V and N entities (including themselves) are adjacent in the knowledge-graph, and if so, it is 1, otherwise, it is 0.
S36: each relationship in the set of relationships W is converted into a D-dimensional vector to construct a relationship vector matrix.
S37: and constructing an entity-relation neighborhood matrix of N.P according to the relation set W, the entity set V and the adjacent condition of the entity and the relation in the knowledge graph. For example, for one entity V in the entity set V, it is determined whether the entity V and the P relationships are adjacent in the knowledge-graph, and if so, it is 1, otherwise, it is 0.
S38: the entity-entity neighborhood matrix generated in step S35 and the entity vector matrix generated in step S34 are fused to generate an entity neighborhood matrix. The purpose is to select K adjacent entities of an entity to form a new set. For example, the entity is acute myocardial infarction, and according to the entity-entity neighborhood matrix, arrhythmia, troponin I and electrocardiogram are obtained from the entities adjacent to the acute myocardial infarction, and the neighborhood set corresponding to the generated entity neighborhood matrix is arrhythmia, troponin I, electrocardiogram and the like. The value of K is determined by setting. In order to reduce the calculation amount, the value of K is appropriately reduced (k×d, k+.n is taken for each entity).
S39: the entity-relationship neighborhood matrix generated in step S37 and the relationship vector matrix generated in step S36 are fused to generate a relationship neighborhood matrix. The purpose is to select K adjacent relations of the entities to form a new set. For example, the entity is acute myocardial infarction, according to the entity-relation neighborhood matrix, the clinical manifestation and the examination and examination of the relation adjacent to the acute myocardial infarction are obtained, and the neighborhood set corresponding to the generated relation neighborhood matrix is the clinical manifestation and the examination and examination. The value of K can be determined by setting. In order to reduce the calculation amount, the value of K may be reduced appropriately (k×d, k+.p for each entity).
S310: the relationship neighborhood matrix generated by the fusion in step S39 is inner-product with the patient feature vector after the nonlinear transformation in step S33, and then summed in the last dimension to generate the patient-relationship score. The vector dimension of the patient-relationship score is sum ([ K, D ] [1, D ], -1) = [ K,1] (-1 represents summation in the last dimension).
S311: the patient-relationship score is inner-product with the entity neighborhood matrix generated by the fusion of step S38 to generate a patient neighborhood matrix. The vector dimension of the patient neighborhood matrix is [ K,1] [ K, D ] = [ K, D ].
S312: the patient neighborhood matrix and the entity neighborhood matrix generated by the fusion of step S38 are added to generate an updated patient neighborhood matrix. The vector dimension of the updated patient neighborhood matrix is [ K, D ] + [ K, D ] = [ K, D ].
S313: and (3) carrying out inner product on the updated patient neighborhood matrix and the patient characteristic vector subjected to nonlinear transformation in the step (S33), and then carrying out one-dimensional summation to generate a diagnosis and treatment scheme vector so as to construct the completed graph neural network model. The dimension of the diagnosis and treatment scheme vector is sum ([ N, D ] [1, D ], -1) = [ N,1].
In the construction of the graph neural network model, the knowledge graph information is skillfully integrated, and the method has the following advantages:
(1) And automatically capturing the high-order structure and semantic information in the knowledge graph by the way that the entity neighborhood matrix and the relation neighborhood matrix participate in the inner product. (2) The fields are fused, and local adjacent structures in each entity are captured and stored, so that a more comprehensive diagnosis and treatment scheme is given. (3) And weighting the neighbors according to the connection relation and the patient score, and representing the semantic information of the knowledge graph.
A graph neural network model building system comprises an acquisition module, a first building module and a second building module. The acquisition module is used for acquiring historical diagnosis and treatment data. The first construction module is used for constructing a knowledge graph triplet according to the entity in the historical diagnosis and treatment data and the predefined relation between every two entities so as to establish the knowledge graph. The second construction module is used for constructing a graph neural network model according to the knowledge graph.
In some examples of the invention, the historical diagnostic data includes structured data and unstructured data, and the graph neural network model building system further includes a first extraction module and a second extraction module. The first extraction module is used for directly extracting entities and relations in the structured data. The second extraction module is used for extracting entities in the unstructured data according to the entity extraction model so as to generate entity data. The second extraction module is also used for extracting the relation between every two entities in the entity data according to the relation extraction model.
A recommendation method of a diagnosis and treatment scheme comprises the following steps:
s40: acquiring diagnosis and treatment entity data of a plurality of patients;
constructing diagnosis and treatment knowledge map triplets according to diagnosis and treatment entities in diagnosis and treatment entity data and a predefined diagnosis and treatment relationship between every two diagnosis and treatment entities so as to establish a plurality of diagnosis and treatment knowledge maps;
processing a plurality of diagnosis and treatment knowledge maps according to a graph neural network model to generate a plurality of diagnosis and treatment scheme vectors, wherein the graph neural network model is constructed by adopting the construction method;
s50: medical entity data of a patient to be recommended is acquired.
And constructing a medical knowledge graph triplet according to the medical entity in the medical entity data and the predefined medical relation between every two medical entities so as to establish a medical knowledge graph.
And processing the medical knowledge graph according to the graph neural network model to generate a medical vector to be recommended.
S60: and calculating the similarity between the medical vector to be recommended and each diagnosis and treatment scheme vector.
S70: and recommending the diagnosis and treatment scheme corresponding to the diagnosis and treatment scheme vector with the similarity value meeting the preset condition according to the similarity.
A system for recommending diagnosis and treatment schemes comprises an acquisition module, a processing module, a third construction module, a calculation module and a selection module. The acquisition module is used for acquiring diagnosis and treatment entity data of a plurality of patients and medical entity data of patients to be recommended. The third construction module is used for constructing diagnosis and treatment knowledge graph triples according to diagnosis and treatment entities in the diagnosis and treatment entity data and the diagnosis and treatment relation predefined between every two diagnosis and treatment entities so as to establish a diagnosis and treatment knowledge graph; the medical knowledge graph three-dimensional structure is used for constructing a medical knowledge graph three-dimensional structure according to the medical entity in the medical entity data and the medical relation predefined between every two medical entities so as to establish a medical knowledge graph. The processing module is used for processing the diagnosis and treatment knowledge maps according to the graph neural network model so as to generate a plurality of diagnosis and treatment scheme vectors. The processing module is also used for processing the medical knowledge graph according to the graph neural network model so as to generate a medical vector to be recommended. The computing module is used for computing the similarity between the medical vector to be recommended and each diagnosis and treatment scheme vector. The selecting module is used for selecting a diagnosis and treatment scheme corresponding to the diagnosis and treatment scheme vector with the similarity value meeting the preset condition according to the similarity.
An electronic device includes a processor and a memory; the processor executes a program corresponding to the executable program code by reading the executable program code stored in the memory, so as to implement the graph neural network model construction method.
A computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements a graph neural network model building method as described above.
A method of recommending a medical regimen, comprising the steps of:
(1) The method comprises the following steps of constructing a graph neural network model:
step 1: historical diagnosis and treatment data of a plurality of patients are obtained.
Step 2: constructing a knowledge graph about diseases based on the historical diagnosis and treatment data obtained in the step 1, comprising the following steps:
step 2.1: the entity type of the entity in which the historic diagnosis and treatment data appears is defined. Specifically, the number of the defined entity types can be up to 500.
Step 2.2: according to the defined entity types, defining the relation between every two entities.
Step 2.3: generating a knowledge graph according to the relation between the entity of which the entity type is defined in the step 2.1 and the entity defined in the step 2.2, wherein the knowledge graph is specifically as follows:
and 2.3.1, if the acquired historical diagnosis and treatment data are structured data, directly extracting the entity in the structured data and the relation between every two entities.
Step 2.3.2, if the acquired historical diagnosis and treatment data is unstructured data, extracting entities and relations from the unstructured data comprises the following steps:
1) Unstructured data is acquired.
2) The entity type of the entity in the portion of unstructured data is manually annotated to generate entity type training data.
3) The bilstm+crf model is trained from the entity type training data to generate the entity extraction model.
4) Predicting entity types of entities in the unstructured data according to the entity extraction model to generate predicted data, wherein the predicted unstructured data is generated by deleting unstructured data corresponding to entity type training data in the unstructured data.
5) And manually checking whether the entity in the predicted data is correctly corresponding to the entity type, and adding the correctly corresponding predicted data into the entity type training data to generate updated entity type training data.
6) Determining that all entities and entity types in the predicted data are correctly corresponding; if not, replacing the entity type training data by the updated entity type training data, and repeating the steps 3) -5); if yes, extracting the entity in the last generated updated entity type training data and the predicted data to generate entity data.
7) Sentences in the entity data are divided into sentences according to periods.
8) And finding out a possible combination set of every two entities in the sentence of the clause.
9) And manually labeling the relation between every two entities in part of entity data according to the combined set to generate relation training data.
Specifically, the notation 1 or 0,1 indicates a relationship, and 0 indicates no relationship.
10 Training the PCNN model based on the relationship training data to generate a relationship extraction model.
11 And predicting the relation between every two entities in the predicted entity data according to the relation extraction model to generate predicted relation data, wherein the predicted entity data is generated by deleting entity data corresponding to relation training data in the entity data.
12 Manually checking whether the relation between every two entities in the predicted relation data is correctly corresponding, and adding the correctly corresponding predicted relation data into the relation training data to generate updated relation training data.
13 Determining whether the relationships between all the two entities in the predicted relationship data are correctly corresponding, if not, replacing the relationship training data by the updated relationship training data, and repeating the steps 10) -12), if so, extracting the relationship between the updated relationship training data and the predicted relationship data which are generated last time.
Step 2.3.2: and constructing a knowledge-graph triplet based on the extracted entity and the relation between every two entities, and constructing the knowledge-graph triplet to form a knowledge graph.
Step 3: constructing a graph neural network model according to the knowledge graph construction, comprising the following steps:
step 3.1: a patient set, an entity set, and a relationship set are constructed.
For example, from the acquired diagnosis and treatment data of M patients, a patient set U, u= { U1, U2, …, uM }, where U represents each patient, is constructed.
From all the entities that M patients have, i.e. N entities, a set of entities V, v= { V1, V2, …, vN }, where V represents each entity, is constructed. For example, a certain drug (nitroglycerin injection), a certain disease (coronary heart disease) and a certain examination (electrocardiogram) all constitute one entity.
From all the relationships that M patients have, i.e., P relationships, a relationship set W, w= { W1, W2, …, wP }, where W represents each relationship, is constructed. For example, the relationship "surgical treatment" between the entities of "acute lower wall posterior wall right ventricular myocardial infarction" and "coronary stent implantation" is one relationship.
Step 3.2: and constructing M-N patient-entity matrixes according to the interaction conditions of each patient and each entity in the patient set U and the entity set V. For example, in the patient entity interaction matrix, it is determined whether there is an interaction relationship between the patient and the entity, and if there is an interaction, it is 1, otherwise it is marked as 0. Each row N-dimension in the patient-entity matrix represents a patient feature vector. For example, one row of data of the patient-entity matrix is extracted to form the patient feature vector.
Step 3.3: the patient feature vector is converted to a D-dimensional patient feature vector by a nonlinear transformation. The dimensions of the patient feature vector are [1, d ], i.e., the dimensions in which the marker in the patient feature vector marked 0 is eliminated.
Step 3.4: each entity in the set of entities V is converted into a D-dimensional vector to construct an entity vector matrix.
Step 3.5: and constructing an N-N entity-entity neighborhood matrix according to the entity set V and the adjacent condition of every two entities in the knowledge graph. For example, for one entity V in the entity set V, it is determined whether the entity V and N entities (including themselves) are adjacent in the knowledge-graph, and if so, it is 1, otherwise, it is 0.
Step 3.6: each relationship in the set of relationships W is converted into a D-dimensional vector to construct a relationship vector matrix.
Step 3.7: and constructing an entity-relation neighborhood matrix of N.P according to the relation set W, the entity set V and the adjacent condition of the entity and the relation in the knowledge graph. For example, for one entity V in the entity set V, it is determined whether the entity V and the P relationships are adjacent in the knowledge-graph, and if so, it is 1, otherwise, it is 0.
Step 3.8: and (3) fusing the entity-entity neighborhood matrix generated in the step (3.5) and the entity vector matrix generated in the step (3.4) to generate an entity neighborhood matrix. The purpose is to select K adjacent entities of an entity to form a new set. For example, the entity is acute myocardial infarction, and according to the entity-entity neighborhood matrix, arrhythmia, troponin I and electrocardiogram are obtained from the entities adjacent to the acute myocardial infarction, and the neighborhood set corresponding to the generated entity neighborhood matrix is arrhythmia, troponin I, electrocardiogram and the like. The value of K is determined by setting. In order to reduce the calculation amount, the value of K is appropriately reduced (k×d, k+.n is taken for each entity).
Step 3.9: and (3) fusing the entity-relationship neighborhood matrix generated in the step (3.7) and the relationship vector matrix generated in the step (3.6) to generate a relationship neighborhood matrix. The purpose is to select K adjacent relations of the entities to form a new set. For example, the entity is acute myocardial infarction, according to the entity-relation neighborhood matrix, the clinical manifestation and the examination and examination of the relation adjacent to the acute myocardial infarction are obtained, and the neighborhood set corresponding to the generated relation neighborhood matrix is the clinical manifestation and the examination and examination. The value of K can be determined by setting. In order to reduce the calculation amount, the value of K may be reduced appropriately (k×d, k+.p for each entity).
Step 3.10: and (3) carrying out inner product on the relation neighborhood matrix generated by fusion in the step (3.9) and the patient characteristic vector subjected to nonlinear transformation in the step (3.3), and then summing according to the last dimension to generate a patient-relation score. The vector dimension of the patient-relationship score is sum ([ K, D ] [1, D ], -1) = [ K,1] (-1 represents summation in the last dimension).
Step 3.11: and (3) performing inner product on the patient-relation score and the entity neighborhood matrix generated by fusing in the step (3.8) to generate a patient neighborhood matrix. The vector dimension of the patient neighborhood matrix is [ K,1] [ K, D ] = [ K, D ].
Step 3.12: and adding the patient neighborhood matrix and the entity neighborhood matrix generated by the fusion in the step 3.8 to generate an updated patient neighborhood matrix. The vector dimension of the updated patient neighborhood matrix is [ K, D ] + [ K, D ] = [ K, D ].
Step 3.13: and (3) carrying out inner product on the updated patient neighborhood matrix and the patient characteristic vector subjected to nonlinear transformation in the step (3.3), and then carrying out one-dimensional summation to generate a diagnosis and treatment scheme vector so as to construct the completed graph neural network model. The dimension of the diagnosis and treatment scheme vector is sum ([ N, D ] [1, D ], -1) = [ N,1].
Step 4: and predicting by using the constructed graph neural network model.
Step 4.1: a plurality of treatment plan vectors for a plurality of patients is acquired.
Step 4.1.1: the method comprises the steps of obtaining a plurality of diagnosis and treatment scheme vectors of a plurality of patients, and specifically comprises the following steps:
step 4.1.2: according to diagnosis and treatment entity in the diagnosis and treatment entity data and the diagnosis and treatment relation predefined between every two diagnosis and treatment entities, constructing diagnosis and treatment knowledge map triplets so as to establish a plurality of diagnosis and treatment knowledge maps.
Step 4.1.3: and processing the multiple diagnosis and treatment knowledge maps according to the graph neural network model to generate multiple diagnosis and treatment scheme vectors.
Step 4.2: the medical vector to be recommended of the patient to be recommended is obtained, and the medical vector to be recommended of the patient to be recommended is specifically as follows:
step 4.2.1: medical entity data of a patient to be recommended is acquired.
Step 4.2.2: and constructing a medical knowledge graph triplet according to the medical entity in the medical entity data and the predefined medical relation between every two medical entities so as to establish a medical knowledge graph.
Step 4.2.3: and processing the medical knowledge graph according to the graph neural network model to generate a medical vector to be recommended.
Step 4.3: and calculating the similarity between the medical vector to be recommended and each diagnosis and treatment scheme vector by adopting a cosine similarity function.
Step 4.4: and recommending the diagnosis and treatment scheme corresponding to the diagnosis and treatment scheme vector with the maximum similarity value according to the similarity.
In the description of the present specification, reference to the terms "some embodiments," "optionally," "further," or "particular embodiments," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, the different embodiments or examples described in this specification and the features of the different embodiments or examples may be combined and combined by those skilled in the art without contradiction.
While embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and not to be construed as limiting the invention, and that variations, modifications, alternatives and variations may be made to the above embodiments by one of ordinary skill in the art within the scope of the invention.
Claims (10)
1. The method for constructing the graph neural network model is characterized by comprising the following steps of:
acquiring historical diagnosis and treatment data of a plurality of patients;
according to the predefined relation between the entity and every two entities in the historical diagnosis and treatment data, constructing a knowledge graph triplet to establish the knowledge graph; the method comprises the steps of,
constructing a graph neural network model according to the knowledge graph, wherein constructing the graph neural network model according to the knowledge graph comprises the following steps:
constructing a patient set, an entity set and a relation set;
constructing a patient-entity matrix according to the interaction conditions of each patient and each entity in the patient set and the entity set;
constructing a patient feature vector according to the patient-entity matrix;
non-linearly transforming the patient feature vector into a D-dimensional patient feature vector, D being a positive integer;
converting each entity in the entity set into a D-dimensional vector to construct an entity vector matrix;
converting each relationship in the relationship set into a D-dimensional vector to construct a relationship vector matrix;
constructing an entity-entity neighborhood matrix according to the entity set and the adjacent situation of every two entities in the knowledge graph;
constructing an entity-relationship neighborhood matrix according to the relationship set, the entity set and the adjacent conditions of the entity and the relationship in the knowledge graph;
Fusing the entity-entity neighborhood matrix and the entity vector matrix to generate an entity neighborhood matrix;
fusing the entity-relationship neighborhood matrix and the relationship vector matrix to generate a relationship neighborhood matrix;
performing inner product and one-dimensional summation on the relation neighborhood matrix and the D-dimensional patient feature vector to generate a patient-relation score;
performing inner product on the patient-relation score and the entity neighborhood matrix to generate a patient neighborhood matrix;
adding the patient neighborhood matrix and the entity neighborhood matrix to generate an updated patient neighborhood matrix;
and carrying out inner product and one-dimensional summation on the updated patient neighborhood matrix and the D-dimensional patient feature vector to generate a diagnosis and treatment scheme vector so as to construct a completed graph neural network model.
2. The method for constructing a graph neural network model according to claim 1, wherein the historical diagnosis and treatment data comprises structured data and unstructured data, and the construction of the knowledge-graph triples comprises the following steps:
directly extracting entities and relations in the structured data;
extracting entities in the unstructured data according to an entity extraction model to generate entity data;
Extracting the relation between every two entities in the entity data according to a relation extraction model; the method comprises the steps of,
and constructing the knowledge graph triplet according to the entity and the relation.
3. The method for constructing a neural network model according to claim 2, wherein extracting the entity in the unstructured data using an entity extraction model comprises the steps of:
1) Obtaining the unstructured data;
2) Labeling entity types of entities in the unstructured data to generate entity type training data;
3) Training a preset entity learning model according to the entity type training data to generate the entity extraction model;
4) Predicting entity types of entities in the unstructured data according to the entity extraction model to generate predicted data, wherein the predicted unstructured data is generated by deleting unstructured data corresponding to the entity type training data in the unstructured data;
5) Checking whether the entity in the predicted data is correctly corresponding to the entity type, and adding the correctly corresponding predicted data into the entity type training data to generate updated entity type training data;
6) Determining whether entity types of all entities in the prediction data are correctly corresponding; if not, replacing the entity type training data by the updated entity type training data, and repeating the steps 3) -5); and if so, extracting the entity in the last generated updated entity type training data and the predicted data to generate the entity data.
4. A graph neural network model building method according to claim 3, characterized in that extracting the relationship from the entity data comprises the steps of:
7) Dividing sentences in the entity data according to periods;
8) Finding out a possible combination set of every two entities in sentences of the clause;
9) Marking the relation between every two entities in part of the entity data according to the combined set to generate relation training data;
10 Training a preset relationship learning model according to the relationship training data to generate the relationship extraction model;
11 Predicting the relation between every two entities in predicted entity data according to the relation extraction model to generate predicted relation data, wherein the predicted entity data is generated by deleting entity data corresponding to the relation training data in the entity data;
12 Checking whether the relation between every two entities in the predicted relation data is correctly corresponding, and adding the correctly corresponding predicted relation data into the relation training data to generate updated relation training data; the method comprises the steps of,
13 Determining whether the relation between every two entities in the predicted relation data is correctly corresponding; if not, replacing the relation training data by adopting updated relation training data, and repeating the steps 10) -12); and if so, extracting the relationship between the updated relationship training data and the predicted relationship data which are generated last time.
5. A graph neural network model building system, comprising:
the acquisition module is used for acquiring historical diagnosis and treatment data;
the first construction module is used for constructing a knowledge spectrum triplet according to the entity in the historical diagnosis and treatment data and the predefined relation between every two entities so as to establish the knowledge spectrum; the method comprises the steps of,
the second construction module is used for constructing a graph neural network model according to the knowledge graph, wherein the construction of the graph neural network model according to the knowledge graph comprises the following steps:
constructing a patient set, an entity set and a relation set;
constructing a patient-entity matrix according to the interaction conditions of each patient and each entity in the patient set and the entity set;
Constructing a patient feature vector according to the patient-entity matrix;
non-linearly transforming the patient feature vector into a D-dimensional patient feature vector, D being a positive integer;
converting each entity in the entity set into a D-dimensional vector to construct an entity vector matrix;
converting each relationship in the relationship set into a D-dimensional vector to construct a relationship vector matrix;
constructing an entity-entity neighborhood matrix according to the entity set and the adjacent situation of every two entities in the knowledge graph;
constructing an entity-relationship neighborhood matrix according to the relationship set, the entity set and the adjacent conditions of the entity and the relationship in the knowledge graph;
fusing the entity-entity neighborhood matrix and the entity vector matrix to generate an entity neighborhood matrix;
fusing the entity-relationship neighborhood matrix and the relationship vector matrix to generate a relationship neighborhood matrix;
performing inner product and one-dimensional summation on the relation neighborhood matrix and the D-dimensional patient feature vector to generate a patient-relation score;
performing inner product on the patient-relation score and the entity neighborhood matrix to generate a patient neighborhood matrix;
adding the patient neighborhood matrix and the entity neighborhood matrix to generate an updated patient neighborhood matrix;
And carrying out inner product and one-dimensional summation on the updated patient neighborhood matrix and the D-dimensional patient feature vector to generate a diagnosis and treatment scheme vector so as to construct a completed graph neural network model.
6. The graphic neural network model building system of claim 5, wherein the historical diagnostic data includes structured data and unstructured data, the graphic neural network model building system further comprising:
the first extraction module is used for directly extracting entities and relations in the structured data; the method comprises the steps of,
the second extraction module is used for extracting entities in the unstructured data according to the entity extraction model so as to generate entity data; and extracting the relation between every two entities in the entity data according to a relation extraction model.
7. The recommendation method of the diagnosis and treatment scheme is characterized by comprising the following steps of:
acquiring diagnosis and treatment entity data of a plurality of patients;
constructing diagnosis and treatment knowledge map triplets according to diagnosis and treatment entities in the diagnosis and treatment entity data and a predefined diagnosis and treatment relationship between every two diagnosis and treatment entities so as to establish a plurality of diagnosis and treatment knowledge maps;
processing a plurality of said diagnosis and treatment knowledge maps according to a graph neural network model, which is constructed according to the construction method as claimed in any one of claims 1 to 4, to generate a plurality of diagnosis and treatment plan vectors; the method comprises the steps of,
Acquiring medical entity data of a patient to be recommended;
according to the medical entity in the medical entity data and the predefined medical relation between every two medical entities, constructing a medical knowledge graph triplet to establish a medical knowledge graph;
processing the medical knowledge graph according to the graph neural network model to generate a medical vector to be recommended;
calculating the similarity between the medical vector to be recommended and each diagnosis and treatment scheme vector; the method comprises the steps of,
and recommending a diagnosis and treatment scheme corresponding to the diagnosis and treatment scheme vector with the similarity value meeting the preset condition according to the similarity.
8. A system for recommending a medical plan, characterized in that it recommends a medical plan by the recommendation method of a medical plan according to claim 7, comprising:
the acquisition module is used for acquiring diagnosis and treatment entity data of a plurality of patients and medical entity data of patients to be recommended;
the third construction module is used for constructing diagnosis and treatment knowledge map triplets according to diagnosis and treatment entities in the diagnosis and treatment entity data and the diagnosis and treatment relation predefined between every two diagnosis and treatment entities so as to establish diagnosis and treatment knowledge maps; the medical knowledge graph triad is used for building a medical knowledge graph according to the medical entity in the medical entity data and the predefined medical relation between every two medical entities so as to build the medical knowledge graph;
The processing module is used for processing a plurality of diagnosis and treatment knowledge maps according to the graph neural network model so as to generate a plurality of diagnosis and treatment scheme vectors; the medical knowledge graph is used for processing the medical knowledge graph according to the graph neural network model to generate a medical vector to be recommended;
the calculation module is used for calculating the similarity between the medical vector to be recommended and each diagnosis and treatment scheme vector; the method comprises the steps of,
and the selecting module is used for selecting the diagnosis and treatment scheme corresponding to the diagnosis and treatment scheme vector with the similarity value meeting the preset condition according to the similarity.
9. An electronic device comprising a processor and a memory; wherein the processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, for implementing the graph neural network model construction method according to any one of claims 1 to 4.
10. A computer-readable storage medium having stored thereon a computer program, which when executed by a processor implements the graph neural network model building method according to any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110690423.0A CN113434692B (en) | 2021-06-22 | 2021-06-22 | Method, system and equipment for constructing graphic neural network model and recommending diagnosis and treatment scheme |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110690423.0A CN113434692B (en) | 2021-06-22 | 2021-06-22 | Method, system and equipment for constructing graphic neural network model and recommending diagnosis and treatment scheme |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113434692A CN113434692A (en) | 2021-09-24 |
| CN113434692B true CN113434692B (en) | 2023-08-01 |
Family
ID=77757120
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110690423.0A Active CN113434692B (en) | 2021-06-22 | 2021-06-22 | Method, system and equipment for constructing graphic neural network model and recommending diagnosis and treatment scheme |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113434692B (en) |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109753570A (en) * | 2019-01-11 | 2019-05-14 | 中山大学 | A kind of scene map vectorization method based on Horn logical AND figure neural network |
| WO2020068831A1 (en) * | 2018-09-26 | 2020-04-02 | Visa International Service Association | Dynamic graph representation learning via attention networks |
| CN111488465A (en) * | 2020-04-14 | 2020-08-04 | 税友软件集团股份有限公司 | Knowledge graph construction method and related device |
| CN112037920A (en) * | 2020-08-31 | 2020-12-04 | 康键信息技术（深圳）有限公司 | Medical knowledge map construction method, device, equipment and storage medium |
| CN112151141A (en) * | 2020-09-23 | 2020-12-29 | 康键信息技术（深圳）有限公司 | Medical data processing method, device, computer equipment and storage medium |
| CN112242187A (en) * | 2020-10-26 | 2021-01-19 | 平安科技（深圳）有限公司 | Medical scheme recommendation system and method based on knowledge graph representation learning |
| CN112257841A (en) * | 2020-09-03 | 2021-01-22 | 北京大学 | Data processing method, device and equipment in graph neural network and storage medium |
| CN112542223A (en) * | 2020-12-21 | 2021-03-23 | 西南科技大学 | Semi-supervised learning method for constructing medical knowledge graph from Chinese electronic medical record |
| CN112613602A (en) * | 2020-12-25 | 2021-04-06 | 神行太保智能科技(苏州)有限公司 | Recommendation method and system based on knowledge-aware hypergraph neural network |
| CN112632250A (en) * | 2020-12-23 | 2021-04-09 | 南京航空航天大学 | Question and answer method and system under multi-document scene |
| CN112989134A (en) * | 2021-03-29 | 2021-06-18 | 腾讯科技（深圳）有限公司 | Node relation graph processing method, device, equipment and storage medium |
| CN112989064A (en) * | 2021-03-16 | 2021-06-18 | 重庆理工大学 | Recommendation method for aggregating knowledge graph neural network and self-adaptive attention |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11886971B2 (en) * | 2019-08-15 | 2024-01-30 | Microsoft Technology Licensing, Llc | Multiple-entity-based recommendation system |
-
2021
- 2021-06-22 CN CN202110690423.0A patent/CN113434692B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020068831A1 (en) * | 2018-09-26 | 2020-04-02 | Visa International Service Association | Dynamic graph representation learning via attention networks |
| CN109753570A (en) * | 2019-01-11 | 2019-05-14 | 中山大学 | A kind of scene map vectorization method based on Horn logical AND figure neural network |
| CN111488465A (en) * | 2020-04-14 | 2020-08-04 | 税友软件集团股份有限公司 | Knowledge graph construction method and related device |
| CN112037920A (en) * | 2020-08-31 | 2020-12-04 | 康键信息技术（深圳）有限公司 | Medical knowledge map construction method, device, equipment and storage medium |
| CN112257841A (en) * | 2020-09-03 | 2021-01-22 | 北京大学 | Data processing method, device and equipment in graph neural network and storage medium |
| CN112151141A (en) * | 2020-09-23 | 2020-12-29 | 康键信息技术（深圳）有限公司 | Medical data processing method, device, computer equipment and storage medium |
| CN112242187A (en) * | 2020-10-26 | 2021-01-19 | 平安科技（深圳）有限公司 | Medical scheme recommendation system and method based on knowledge graph representation learning |
| CN112542223A (en) * | 2020-12-21 | 2021-03-23 | 西南科技大学 | Semi-supervised learning method for constructing medical knowledge graph from Chinese electronic medical record |
| CN112632250A (en) * | 2020-12-23 | 2021-04-09 | 南京航空航天大学 | Question and answer method and system under multi-document scene |
| CN112613602A (en) * | 2020-12-25 | 2021-04-06 | 神行太保智能科技(苏州)有限公司 | Recommendation method and system based on knowledge-aware hypergraph neural network |
| CN112989064A (en) * | 2021-03-16 | 2021-06-18 | 重庆理工大学 | Recommendation method for aggregating knowledge graph neural network and self-adaptive attention |
| CN112989134A (en) * | 2021-03-29 | 2021-06-18 | 腾讯科技（深圳）有限公司 | Node relation graph processing method, device, equipment and storage medium |
Non-Patent Citations (4)
| Title |
|---|
| Stability of Graph Neural Networks to Relative Perturbations;Fernando Gama 等;《ICASSP 2020 - 2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)》;9070-9074 * |
| Tianxiang Zhao.GraphSMOTE: Imbalanced Node Classification on Graphs with Graph Neural Networks.《WSDM '21: Proceedings of the 14th ACM International Conference on Web Search and Data Mining》.2021,833-841. * |
| 图神经网络推荐研究进展;吴国栋 等;《智能系统学报 》;14-24 * |
| 基于知识图谱驱动的图神经网络推荐模型;刘欢 等;《计算机应用》;1865-1870 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113434692A (en) | 2021-09-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | Detection and classification of cardiac arrhythmias by a challenge-best deep learning neural network model | |
| Guidi et al. | A machine learning system to improve heart failure patient assistance | |
| Chauhan et al. | Heart disease prediction using evolutionary rule learning | |
| US11915828B2 (en) | System and method of using machine learning for extraction of symptoms from electronic health records | |
| Schoolman et al. | Computer Use in Diagnosis, Prognosis, and Therapy: Computers are used to simulate the reasoning of expert clinicians. | |
| Yan et al. | Selecting critical clinical features for heart diseases diagnosis with a real-coded genetic algorithm | |
| Arvindhan et al. | A review of challenges and opportunities in machine learning for healthcare | |
| Mlakar et al. | Mining telemonitored physiological data and patient-reported outcomes of congestive heart failure patients | |
| Chou et al. | Extracting drug utilization knowledge using self-organizing map and rough set theory | |
| CN112542242A (en) | Data transformation/symptom scoring | |
| CN115565660A (en) | Medical image data sharing management system based on feature recognition | |
| CN112259260B (en) | Intelligent medical question-answering method, system and device based on intelligent wearable equipment | |
| JPWO2019220833A1 (en) | Diagnostic support system, diagnostic support device and diagnostic support method | |
| Bisercic et al. | Interpretable medical diagnostics with structured data extraction by large language models | |
| Siontis et al. | ChatGPT hallucinating: can it get any more humanlike? | |
| CN113434692B (en) | Method, system and equipment for constructing graphic neural network model and recommending diagnosis and treatment scheme | |
| Sarp et al. | Artificial intelligence-powered chronic wound management system: Towards human digital twins | |
| Andry et al. | Electronic health record to predict a heart attack used data mining with Naïve Bayes method | |
| Subasi et al. | Digital twins in healthcare and biomedicine | |
| Mehrankia et al. | Prediction of heart attacks using biological signals based on recurrent GMDH neural network | |
| Strobel et al. | Healthcare in the Era of Digital twins: towards a Domain-Specific Taxonomy. | |
| Ganipineni et al. | Artificial Intelligence in Cardiovascular Medicine: A Comprehensive Clinical Review | |
| Vatian et al. | Retrieving and Processing Information from Clinical Algorithm via Formal Concept Analysis | |
| CN117009839B (en) | Patient clustering method and device based on heterogeneous hypergraph neural network | |
| Oliveira et al. | Explaining end-to-end ECG automated diagnosis using contextual features |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |